摘要
橡胶制品表面缺陷会极大地影响产品的性能、安全性和可靠性。传统的基于人的视觉检测准确率低且耗时长,现有的机器视觉检测技术主要依靠人工完成,无法有效精准对橡胶制品缺陷检测。针对橡胶制品中各种缺陷检测识别问题,采用基于深度学习模型的多尺度缺陷检测方法,使用橡胶制品表面缺陷图像对该网络进行了训练和评估。研究结果显示,所建立的检测模型对橡胶制品凹坑检测准确率较高,平均精度为92.7%,可以有效检测橡胶制品小缺陷,且基于深度学习的神经网络模型的初始总损失相对较小,在100~150个历时之间趋于稳定。
Surface defects in rubber products can greatly affect the performance,safety,and reliability of the product.Traditional human based visual inspection has low accuracy and long time consumption.However,existing machine vision inspection technologies mainly rely on manual work and cannot effectively and accurately detect defects in rubber products.Therefore,in response to various defect detection and recognition issues in rubber products,a multi-scale defect detection method based on depth models is adopted.The network was trained and evaluated using surface defect images of rubber products.The research results show that the detection model established in this article has a high accuracy in detecting pits in rubber products,with an average accuracy of 92.7%.It can effectively detect small defects in rubber products.Moreover,the initial total loss of the neural network model based on deep learning is relatively small,and tends to stabilize between 100 and 150 durations.
作者
朱世元
方世鹏
ZHU Shiyuan;FANG Shipeng(Yan’an University,Yan’an 716000,Shaanxi China)
出处
《粘接》
CAS
2023年第7期26-29,共4页
Adhesion
关键词
深度学习
橡胶制品
缺陷
识别
技术研究
deep learning
rubber products
defects
recognition
technical research
